Learning with Intelligent Agents: A Longitudinal Study

Aaron H. Doering and George Veletsianos

Abstract

Previous empirical research is not clear on whether intelligent agent integration correlates with learning gains. A number of researchers found no increased learning performance with the use of agents (Baylor, 2002; Mayer et al., 2003; Craig et al., 2002); others found learner performance could be improved when agents are employed in a well-structured learning environment (Atkinson, 2002; Moreno et al., 2001), even when the agent is simply a disembodied head lacking natural facial expressions (Cole et al, 2003). In this study, we are attempting to engage participants in a dynamic discourse with the agent over a period of four weeks. Such a long-term agent evaluation is important because to date there are no studies on the long-term effects of agent enhanced learning environments. Gulz (2004) posits that, “A significant shortcoming of current evidence is that most studies leave us without knowledge of what happens when learners are involved in repeated interactions with social characters over a longer period of time. As a result we know little about potential effects of character enhancement of computerized learning environments in ecologically valid contexts.”

Current teaching practices emphasize the importance of collaboration (Bransford et al., 1999) and active participation (Jonassen, 2000). We hypothesize that intelligent agents can improve learning because of their ability to be motivating, engaging, and to interact socially with the learner on a one-to-one basis. They allow the learner to actively participate in the learning process, make associations with previous knowledge, and engage in meaningful learning.

We examine the development of an online portfolio system with the assistance of male or female intelligent agents. The agents are characterized by reasoning, reactivity, and emotion and can converse via speech and text with students on matters concerning the design and development of their electronic portfolio. Data showed intelligent agents provided learner motivation, support, and encouragement in the development of their portfolio. However, participants also requested more comprehensive answers to their questions. Furthermore, the agents provided an “online friend” at all times that went beyond the development of their assignment to discussions on a social level.

About the Authors

Aaron Doering is an assistant professor in Learning Technologies in the Department of Curriculum and Instruction at the University of Minnesota. Doering's teaching and research interests relate to the development of effective distance learning environments, technology integration in K-12 preservice and inservice settings, and the innovative use of technology to support teaching and learning.

George Veletsianos is a doctoral student at the University of Minnesota's Learning Technologies program. His research interests are multidisciplinary and revolve around aesthetics, human-computer interaction, social psychology, persuasion, and emotions. He holds a BA in Economics and Computer Science and an MA in Learning Technologies. More information about George can be found at DeliciousPixels.com